[libcamera-devel] ScalerCrop control and camera modes

[Jacopo Mondi]

Hi Dave, Naush

On Wed, Jan 11, 2023 at 04:24:24PM +0000, David Plowman wrote:
> Indeed, I think pretty much all our controls are handled by IPAs so
> the simple "don't do anything" means they get preserved automatically.

Ah, see ? I was not expecting that.

Looking at the RkISP1 implementation for, in example, awb and agc it
seems at IPA::configure() time the algorithms are reset to their
default value, while the RPi IPA only do so at "firstRun_" time.

Interesting as a discussion what application would expect to happen to
controls at configure() time..

> ScalerCrop is a bit different in that the PH deals with it.
>
> To be honest, I can see an argument that says "resetting it when you
> reconfigure" is a simple behaviour that is easily understood. If you
> want to preserve it, you need to pass a value with your call to start.
> I possibly have a slight preference for this because (a) it's easily
> understood and (b) it's what the code does now (!), but I feel we
> should probably make a conscious decision.
>

FWIW, since I was considering all controls to be reset at configure,
the current behavior matches my expectations :)

> David
>
> On Wed, 11 Jan 2023 at 16:07, Naushir Patuck <naush at raspberrypi.com> wrote:
> >
> > Hi Jacopo,
> >
> >
> > On Wed, 11 Jan 2023 at 15:55, Jacopo Mondi via libcamera-devel <libcamera-devel at lists.libcamera.org> wrote:
> >>
> >> Hi David
> >>
> >> On Wed, Jan 11, 2023 at 01:27:31PM +0000, David Plowman via libcamera-devel wrote:
> >> > Hi everyone
> >> >
> >> > A little question about what happens to the ScalerCrop control when we
> >> > switch between camera modes (i.e. call configure). Paying particular
> >> > attention to the case when the two camera modes have different sensor
> >> > crops (and therefore a different FoV), should the ScalerCrop be
> >> > preserved or not?
> >>
> >> How is a Control "preserved" between streaming sessions ?
> >>
> >> Are you wondering if the ScalerCrop set by an application during a
> >> streaming session should be retained by the pipeline handler after a
> >> [stop - configure] sequence and re-applied at start() time without
> >> application's intervention ?
> >>
> >> If that's the case, are there other controls preserved between
> >> reconfigurations or do you have any in mind ?
> >
> >
> > Most controls destined for the IPA currently preserve value between
> > reconfigurations.  For example, ExposureTime, AnalogueGain, Brightness,
> > Contrast, ColourGains, etc. all apply the same settings regardless of mode.  In
> > fact, I cannot think of a control (apart from ScalerCrop) that is not preserved
> > between reconfigurations!
> >
> > Regards,
> > Naush
> >
> >>
> >> >
> >> > I think we could go either way.
> >> >
> >> > If the ScalerCrop is preserved that's fine, but you have to handle the
> >> > case where the previous ScalerCrop is not entirely available in the
> >> > new mode. You also have to consider how an application would reset the
> >> > ScalerCrop so as to switch from a mode with a limited FoV to one with
> >> > a larger FoV.
> >> >
> >> > If the ScalerCrop isn't preserved that's fine too, but you have to
> >> > consider how an application would preserve the ScalerCrop if it wanted
> >> > to.
> >> >
> >> > In many respects I don't think there's much difference, applications
> >> > would (sometimes) have to specify the ScalerCrop they want when they
> >> > call start. But which do we think is the most desirable behaviour? On
> >> > the Pi, by not having thought about it, we are implementing the second
> >> > alternative (ScalerCrop not preserved).
> >> >
> >> > Thoughts?
> >> >
> >> > Thanks!
> >> > David